Memory Sharing among workers in gunicorn using --preload

Question

I have a large model file that I use in my webservice built in Flask and then served through Gunicorn. The folder is strucure is like this:

A.py
Folder_1\
    __init__.py
    B.py

The model is loaded in __init__.py and used in B.py The entry point is A.py that contains @app.routes, etc.

I start A.py with gunicorn and preload the app using --preload option and there are 8 workers.

I am facing 100% CPU utilization on 8 cores; apparently the requests are stuck at app server and not being forwarded to DB. Does the model is also preloaded and made available to all 8 workers, i.e. Is it shared between the worker processes? If not do I have to load the model in A.py so that the model is also preloaded for all workers.

I think the model is being loaded by every worker process and since the model is large, the workers are stuck there.

EDIT 1 : Since I was notified that this might be a duplicate question, I want to clarify I am not asking how python handles shared object. I understand that is possible by using multiprocessing. In my case, I start flask server from gunicorn with 8 workers using --preload option, there are 8 instances of my app running. My question is, since the code was preloaded before workers were forked, gunicorn workers will share the same model object, or they will have a separate copy each.?

Answer 1

My question is, since the code was preloaded before workers were forked, gunicorn workers will share the same model object, or they will have a separate copy each.?

It will be a separate copy.

preloading simply takes advantage of the fact that when you call the operating system's fork() call to create a new process, the OS is able to share unmodified sections of memory between the two processes. By preloading as much code as possible more memory is shared between the processes.

This is simply a behind-the-scenes operating system optimization: from the perspective of each individual python process they have unique copies of each object.